My colleague Bruce Carlisle and I have recently created a new (ish) kind of map. We've taken road data for Newcastle upon Tyne and instead of mapping them geospatially we have mapped them topologically using social network analysis software called Gephi. The connections between roads are maintained, but their location is free to be manipulated.
Gephi allows you to create interesting layouts based on the nature of the connections, nodes and attributes of the network. For this map we used one called Force Atlas 2 which generates a gravity, pulling together nodes (in this case roads) which share connections, and pushing part those which don't.
This process creates a discombobulating map which is unfamiliar upon first appearance, but once you study it places become recognisable. Suburbs and coherent urban areas are grouped together and are located in relation to one another which is logical, but strange at the same time.
Given the roads are correctly connected to one another the map can be used to navigate the city, but time and space are hard to judge. For example, in a test version two streets which are only 300 metres apart 'as the crow flies', appear at different ends of the page.
Edit - 16th June, 2016
The article is available 'online first' here. If you can't access that the text is below:
A Topological Road Map of Newcastle upon Tyne
Jon Swords and Bruce Carlisle, Department of Geography, Northumbria University, UK.
This featured graphic is a topological road map of Newcastle upon Tyne, UK, generated using GIS and social network analysis software using Ordnance Survey data. The map formed part of ‘en_counter’ (2016), an exhibition of mapping work in Newcastle upon tyne.
The map is derived from Ordnance Survey MasterMap Integrated Transport Network data (Ordnance Survey, 2010), provided in a multi-part fully topologically structured link and node format including information about roads (‘RoadLink’), their names (‘Road’) and their connections (‘RoadNode’). Various data manipulations were made using ArcMap GIS and spreadsheet software to transform the data into the flat file format needed by Gephi social network analysis software. Relevant attributes from the ‘RoadNode’ and ‘Road’ items were joined to the ‘RoadLink’ features. The ‘RoadLink’ features were clipped to the boundary of Newcastle Metropolitan District using Ordnance Survey Boundary-Line data (Ordnance Survey, 2015). Only public roads were selected and so excluding alleys and private roads. Some ‘RoadLinks’, mainly segments of roundabouts and slip roads, had no name. A name was added if appropriate, or else the link was deleted. Incidences of the same name being used for different roads were identified and the name was altered to, for example, High Street 1, High Street 2, etc. A series of look-up operations, sorts and filters were used to produce a flat table with each row representing a named road and one of the roads it connects with. The result was data in which connections between roads are maintained, but their location is free to be manipulated.
Gephi allows the manipulation and analysis of social network graphs. We used the Force Atlas 2 algorithm (Jacomy et al., 2014) to generate forces of attraction and repulsion within the network, pulling together nodes (in this case roads) which share connections, and pushing part those which don’t. Refinement of labels, lines and layout was done in Adobe Illustrator.
The final map is accurate in the sense it can be used to navigate the city from road to road, but the lack of geospatial references make it unfamiliar and discombobulating upon viewing. Suburbs and coherent urban areas are grouped together and are located in relation to one another which is logical, but strange at the same time. Topographical scale is lost as map distance is a result of relational connectivity rather than points in physical space.
An approach similar to this has been previously theorized (Park and Yilmaz, 2010), albeit without spatialising algorithms in mind. Statistical analysis of the network can be undertaken including measures of centrality to identify significant roads based on the number of connections. The ever-busy B1600 (Osborne Road-Portland Road-Stoddard Street) has the highest Eigenvector centrality score, and the A1 ring road the highest betweenness centrality. On average each road in the city is connected to three other nodes.
Bonacich P (2007) Some properties of eigenvector centrality Social Networks 29: 555-564.
Messer S, Swords J and Jeffries M (2016) En_counter. Holy Biscuit Gallery Newcastle upon Tyne May 13th – June 2nd.
Jacomy M, Venturini T and Heymann S and Bastian M (2014) Force Atlas 2 a continuous graph layout algorithm for handy network visualization designed for Gephi Software PLoS ONE 9(6): e98679.
Ordnance Survey (2010) OS MasterMap Integrated Transport Network Layer: user guide and technical specification. Available at: https://wwwordnancesurveycouk/docs/user-guides/os-mastermap-itn-layer-user-guidepdf (accessed 19th April 2016).
Ordnance Survey (2015) Ordnance Survey Boundary-Line: product guide and technical specification. Available at: (accessed 19th April 2016).
Park P and Yilmaz A (2010) A Social Network Analysis Approach to Analyze Road Networks. In: ASPRS Annual Conference San Diego, USA, 26-30 April 2010
Scott J (2000) Social network analysis handbook. London: Sage
 Eigenvector centrality measures the centrality of a node taking takes into account the centrality of neighbouring nodes with which it is connected (. Scores are relative, the most important node = 1. (ref). Betweenness centrality measures how often a node appears on paths between nodes in the network. That is, it measures how often a particular node is crossed when tracing the shortest pathway between any two nodes (Scott, 2000)